Initializes go-beast explicitly for a new agent session or environment: classifies the target harness, chooses between planning-only/manual/installer-backed setup, separates canonical core steps from optional harness wiring, and validates that skills, instructions, and optional integrations are actually ready.
Initializes go-beast explicitly for a new agent session or environment: classifies the target harness, chooses between planning-only/manual/installer-backed setup, separates canonical core steps from optional harness wiring, and validates that skills, instructions, and optional integrations are actually ready.
when_to_use
Use when a developer wants to bootstrap go-beast explicitly instead of relying on SessionStart automation, or when hooks are unavailable, untrusted, or undesirable. Invoke before go-mole on an unfamiliar repository and before go-jay or go-swift when baseline skill discovery, instructions, or optional harness wiring must exist first.
go-mule — Agent Initialization & Instrumentation
go-mule carries the pack into place before the rest of the beasts start moving. It turns initialization into an explicit, repeatable workflow instead of hidden background sync.
Quick start
User: "Initialize go-beast for Codex without relying on SessionStart hooks."
→ invoke go-mule
→ classify target and permissions → choose setup path → separate core from optional wiring → validate readiness
Workflow
1. Classify the target and the allowed actions
Identify the initialization surface before proposing any step:
Which agent or harness is being initialized? (Codex, Claude Code, plugin surface, or another agent)
Is the request planning-only, read-only validation, or permission to change files / run commands?
Is this a hookless environment, a hook-capable environment, or a mixed environment with partial support?
Is initialization meant to be session-only, machine-level, or project-local?
Does the user want the sync hook preserved as an option, or explicitly avoided?
If the user has not authorized file changes or command execution, constrain the output to a procedural plan only.
2. Define the canonical core initialization contract
Anchor the workflow in the real go-beast artifacts and separate them by responsibility:
skills/ — canonical skill source
AGENTS.global.md and AGENTS.bootstrap.md — synced global instruction sources
hooks/sync-go-beast-skills.sh — optional SessionStart automation, not the only valid path
Which of these artifacts are relevant to the target harness?
Is bootstrap mode desired, or should the global instructions stay on the standard path?
Is the plugin adapter needed, or is direct file-based skill loading enough?
Which parts are agent-agnostic core setup versus harness-specific wiring?
Produce INITIALIZATION PLAN before suggesting any mutation.
3. Choose the explicit setup path
Select exactly one primary path and justify it:
Planning-only path — no mutation; produce repeatable steps and validation criteria.
Installer-backed path — use scripts/install.mjs when the environment can safely accept explicit installation.
Manual path — link or point the agent at skills/, copy the chosen global instructions file, and leave hooks/plugin wiring optional.
Refresh path — validate an existing installation, fix drift, and report only the missing pieces.
Does the chosen path avoid unnecessary harness coupling?
Can the same path be repeated safely in the same environment?
If the path touches global files, has the user explicitly allowed that scope?
If the path needs more than baseline initialization, should go-jay or go-swift be invoked next instead of folding that work into go-mule?
4. Execute or describe the core setup
The core setup must stay valid even when hooks are absent:
Point the agent at the canonical skills/ directory or run the explicit installer path.
Choose between AGENTS.global.md and AGENTS.bootstrap.md based on the user's desired behavior.
Record whether the environment should rely on manual invocation, installer-backed setup, or later hook automation.
Keep all steps ordered, repeatable, and easy to re-run.
If using scripts/install.mjs, specify the exact mode (interactive, --all, --bootstrap, or --uninstall when relevant).
If using the manual path, name the concrete directories or files that must be pointed at or copied.
If staying planning-only, make the operator sequence explicit enough that another agent could follow it without guessing.
Produce CORE SETUP as the exact ordered sequence or applied changes.
5. Isolate optional harness-specific wiring
Anything AI-surface-specific is additive, never core:
Codex: optional ~/.codex/hooks.json or [hooks] in ~/.codex/config.toml; /hooks trust review after changes
Claude Code: optional ~/.claude/settings.json wiring
Plugin surfaces: optional use of plugins/go-beast/
Bootstrap mode: optional replacement of standard global instructions with AGENTS.bootstrap.md
Which of these steps are safe and relevant for the current harness?
Which ones should remain manual or deferred because the user only asked for explicit initialization?
Does any step require go-jay for instruction editing or go-swift for hook authoring instead of ad-hoc edits here?
Produce OPTIONAL HARNESS STEPS separately from the core path. Never mix them into the baseline sequence.
6. Validate readiness and hand off
Initialization is not done until readiness is proven:
Are the expected skills discoverable from skills/ or the chosen install path?
Is the correct global instruction file active?
If optional hook wiring was touched, has the trust/review step been completed?
Is there any drift between the explicit path and the optional sync-hook path?
Which beast should run next: go-mole, go-hawk, go-jay, or go-swift?
Produce:
VALIDATION REPORT
NEXT BEAST HANDOFF
If readiness cannot be proven, mark the setup blocked and name the exact blocker.
Rules
Do not require SessionStart hooks. Hook automation is one valid path, not the baseline contract.
Keep the core setup agent-agnostic. Put Codex-, Claude-, or plugin-specific wiring only under OPTIONAL HARNESS STEPS.
Do not edit global instructions, hook config, or symlinks unless the user explicitly authorizes file changes or command execution.
If initialization requires new or changed context files, invoke go-jay after the baseline setup. If it requires new or changed hook automation, invoke go-swift.
Do not claim initialization is active until validation proves skill discovery, instruction loading, and any requested optional wiring are actually ready.
Output
INITIALIZATION PLAN — target harness, allowed actions, chosen explicit setup path, and bootstrap mode decision
CORE SETUP — repeatable core initialization steps or applied changes for skills/, instruction files, and the installer path
